Online Junction Temperature Estimation of Power Semiconductor Devices using Neural Network and Model-Based Design

This paper presents an approach for estimating the junction temperature of semiconductor devices using neural networks. Our approach is based on the use of machine learning techniques to model the relationship between the device's electrical and thermal properties without using additional electrical or thermal sensors. Also in this work, the change in active load resistance under the influence of temperature is taken into account. As features for the neural network, the components of power losses which are characteristic for semiconductor devices are used. Also, physics-based data augmentation is used to expand the input data space for the training of the neural network in order to improve the accuracy of the model. An optimal set of features is found to provide an acceptable accuracy of temperature estimation. The results of our study demonstrate the effectiveness of this approach in providing accurate and inline junction temperature estimation, and its implications for the design and operation of semiconductor-based systems.